path: root/lib/net.c

#include "log.h"
#include "arena.h"
#include "macros.h"
#include "sha1.h"
#include "net.h"

#include <libgen.h> /* for dirname/basename */
#include <curl/curl.h>
#include <jansson.h>
#include <errno.h>
#include <string.h>

/* TODO: also support etag/if-not-match rather than just last-modified/if-not-modified */

/* memory usage is about 3 * strlen(path) -- what da heeeellll */
static char *derive_meta_path(vl_arena *arena, const char *path)
{
  char *base = basename(vl_arena_strdup(arena, path));
  char *dname = dirname(vl_arena_strdup(arena, path));
  return vl_arena_sprintf(arena, "%s/.%s.meta", dname, base);
}

#define NET_BUFSIZE (4096)

static int hash_file(const char *fname, vl_sha1 ohash)
{
  FILE *file;
  unsigned char data[NET_BUFSIZE];
  size_t nread;
  int ret = -1;
  vl_sha1_st hashst;

  file = fopen(fname, "rb");

  if (!file) {
    vl_debug("failed to hash file %s: %s", fname, strerror(errno));
    return -1;
  }

  vl_sha1_init(&hashst);
  while ((nread = fread(data, 1, NET_BUFSIZE, file)) > 0) {
    vl_sha1_update(&hashst, data, nread);
  }

  if (ferror(file)) {
    vl_debug("failed to read file %s :(", fname);
    goto cleanup;
  }

  vl_sha1_finalize(&hashst, ohash);

  ret = 0;

cleanup:
  fclose(file);
  return ret;
}

static int check_cache_consistent(vl_arena *arena, const char *path, const char *meta_path, struct curl_slist **headers)
{
  json_error_t jerr;
  int ret = -1;
  json_t *j = json_load_file(meta_path, 0, &jerr);
  char *header = NULL;

  vl_trace("checking meta file %s", meta_path);

  if (!j) {
    vl_debug("failed to load json meta %s(%d:%d:%d): %s", jerr.source, jerr.line, jerr.column, jerr.position, jerr.text);
    goto cleanup;
  }

  const char *lm;
  const char *sha1_hex;
  size_t sha1_len;

  if (json_unpack_ex(j, &jerr, JSON_STRICT, "{s:s, s:s%}", "lm", &lm, "sha1", &sha1_hex, &sha1_len) < 0) {
    vl_debug("failed to unpack json meta %s: %s", meta_path, jerr.text);
    goto cleanup;
  }

  if (sha1_len != VL_SHA1_DIGEST_HEX_STRLEN) {
    vl_debug("failed to read json meta %s: invalid sha1 digest (is %zu chars long, expected %u)", meta_path, sha1_len, VL_SHA1_DIGEST_HEX_STRLEN);
    goto cleanup;
  }

  vl_sha1 hashgot, hashexp;
  if (vl_sha1_decode(hashexp, sha1_hex) < 0) {
    vl_debug("failed to read json meta %s: invalid sha1 digest (bad format)", meta_path);
    goto cleanup;
  }

  if (hash_file(path, hashgot) < 0) {
    goto cleanup;
  }

  if (memcmp(hashexp, hashgot, sizeof(vl_sha1)) != 0) {
#ifdef LOG_DEBUG_ENABLED
    char hash_exp_hex[VL_SHA1_DIGEST_HEX_STRLEN + 1];
    char hash_got_hex[VL_SHA1_DIGEST_HEX_STRLEN + 1];

    hash_exp_hex[VL_SHA1_DIGEST_HEX_STRLEN] = 0;
    hash_got_hex[VL_SHA1_DIGEST_HEX_STRLEN] = 0;

    vl_sha1_encode(hashexp, hash_exp_hex);
    vl_sha1_encode(hashgot, hash_got_hex);

    vl_debug("file %s tampered on disk (sha1 expect %s, got %s)", path, hash_exp_hex, hash_got_hex);
#endif

    goto cleanup;
  }

  header = vl_arena_sprintf(arena, "If-Modified-Since: %s", lm);
  *headers = curl_slist_append(*headers, header);
  ret = 0;

cleanup:
  if (j) json_decref(j);
  return ret;
}

static void write_transfer_meta(CURL *easy, vl_sha1 ohash, const char *meta_path)
{
  json_t *metaj = NULL;
  struct curl_header *hdr;
  CURLHcode hcode;
  char hash_hex[VL_SHA1_DIGEST_HEX_STRLEN + 1];

  vl_trace("writing transfer meta to %s", meta_path);

  hash_hex[VL_SHA1_DIGEST_HEX_STRLEN] = '\0';

  hcode = curl_easy_header(easy, "Last-Modified", 0, CURLH_HEADER, -1, &hdr);
  if (hcode != CURLHE_OK) {
    vl_debug("Not writing meta %s: curl_easy_header(Last-Modified): %u", meta_path, hcode);
    goto cleanup;
  }

  vl_sha1_encode(ohash, hash_hex);
  metaj = json_pack("{s:s, s:s}", "lm", hdr->value, "sha1", hash_hex);
  if (!metaj) {
    vl_debug("Not writing meta %s: json_pack returned NULL (weird)", meta_path);
    goto cleanup;
  }

  if (json_dump_file(metaj, meta_path, JSON_COMPACT) < 0) {
    vl_debug("Failed writing meta to %s: json_dump_file", meta_path);
    goto cleanup;
  }

cleanup:
  if (metaj) json_decref(metaj);
}

struct write_ctx
{
  const char *opath;
  FILE *ofile;
  vl_sha1_st sha1_state;
};

static size_t handle_write(char *ptr, size_t sz, size_t nmemb, void *user)
{
  struct write_ctx *ctx = user;
  if (sz * nmemb == 0) return 0;

  /* Note that the output file is opened lazily, because we might receive a 304 (Not Modified)
   * response with an empty body (in such a case, this function could be called with nmemb == 0).
   * fopen for writing truncates the file, so we can't open the file in the main function.
   */
  if (!ctx->ofile) {
    ctx->ofile = fopen(ctx->opath, "wb");
    if (!ctx->ofile) {
      vl_warn("Failed to open output file: fopen(%s, wb): %s", ctx->opath, strerror(errno));
      return CURL_WRITEFUNC_ERROR;
    }
  }

  vl_sha1_update(&ctx->sha1_state, ptr, sz * nmemb);
  return fwrite(ptr, sz, nmemb, ctx->ofile);
}

int vl_net_ensure_cached(vl_arena *arena, const char *url, const char *target_path)
{
  char *meta_path = derive_meta_path(arena, target_path);
  int ret = -1;
  char errbuf[CURL_ERROR_SIZE];
  CURLcode ecode;
  CURL *easy = NULL;
  struct curl_slist *headers = NULL;
  struct write_ctx wrctx = { .opath = target_path, .ofile = NULL };
  vl_sha1 ohash;
  long response_code;

  vl_trace("Downloading cached file from %s to %s (%s)", url, target_path, meta_path);

  check_cache_consistent(arena, target_path, meta_path, &headers);

  easy = curl_easy_init();
  if (!easy) {
    vl_warn("Failed to set up CURL handle to download %s!", url);
    goto cleanup;
  }

  errbuf[0] = '\0';

  if ((ecode = curl_easy_setopt(easy, CURLOPT_ERRORBUFFER, errbuf)) != CURLE_OK) {
    vl_warn("curl_easy_setopt(CURLOPT_ERRORBUFFER) failed: %s", curl_easy_strerror(ecode));
    goto cleanup;
  }

#define CHECK(_ex) do {                                                     \
  if ((ecode = (_ex)) != CURLE_OK) {                                        \
    vl_warn("%s failed: %s (%s)", #_ex, curl_easy_strerror(ecode), errbuf); \
    goto cleanup;                                                           \
  }                                                                         \
} while (0)

  CHECK(curl_easy_setopt(easy, CURLOPT_URL, url));
  CHECK(curl_easy_setopt(easy, CURLOPT_USERAGENT, VL_USER_AGENT));
  CHECK(curl_easy_setopt(easy, CURLOPT_HTTPGET, 1L));
  CHECK(curl_easy_setopt(easy, CURLOPT_FOLLOWLOCATION, CURLFOLLOW_ALL));
  CHECK(curl_easy_setopt(easy, CURLOPT_TIMEOUT, 60));

  if (headers) {
    CHECK(curl_easy_setopt(easy, CURLOPT_HTTPHEADER, headers));
  }

  vl_sha1_init(&wrctx.sha1_state);

  CHECK(curl_easy_setopt(easy, CURLOPT_WRITEFUNCTION, &handle_write));
  CHECK(curl_easy_setopt(easy, CURLOPT_WRITEDATA, &wrctx));

  CHECK(curl_easy_perform(easy));

  CHECK(curl_easy_getinfo(easy, CURLINFO_RESPONSE_CODE, &response_code));

#undef CHECK

  if (response_code == 304) {
    vl_info("Downloaded file %s not modified.", target_path);
    ret = 0;
    goto cleanup;
  } else if (response_code / 100 == 2) {
    vl_trace("download %s success: %ld", url, response_code);
  } else {
    vl_warn("Bad HTTP response code %ld downloading %s", response_code, url);
    goto cleanup;
  }

  if (wrctx.ofile) {
    fclose(wrctx.ofile);
    wrctx.ofile = NULL;
  }

  vl_sha1_finalize(&wrctx.sha1_state, ohash);
  write_transfer_meta(easy, ohash, meta_path);

  ret = 0;

cleanup:
  if (easy) curl_easy_cleanup(easy);
  if (headers) curl_slist_free_all(headers);
  if (wrctx.ofile) {
    if (ret < 0 && remove(target_path) < 0) {
      vl_debug("... failed to clean up after failed download: unlink(%s): %s", target_path, strerror(errno));
    }

    fclose(wrctx.ofile);
  }

  return ret;
}